OpenAI resolves issue around em-dash patterns in generated text

OpenAI CEO Sam Altman said that ChatGPT will now listen when asked not to use em dashes.

In recent months, people online began treating the punctuation mark (—) as a giveaway that a text was written by an AI. The symbol has indeed appeared more often in student essays and corporate materials.

In reality, em-dashes have long been standard for anyone who works with text professionally or follows formal style guides. Since plenty of pre-AI writing on the internet used them, ChatGPT naturally picked up the habit during training.

The issue was that even when users explicitly asked the model to avoid em-dashes, ChatGPT would ignore the request. According to Altman, this has now been fixed. To disable em-dash usage, the preference must be set in the model’s personalization settings.